SUMMARY: SCSI Reservation conflict with mpxio and SVM metasets

From: Tim Wright <tim.dormouse_at_googlemail.com>
Date: Sun May 14 2006 - 11:40:34 EDT
After an extended period of going back and forth with Sun, they
eventually decided that as we're using non Sun QLogic cards, they
weren't prepared to try and reproduce the issue we were having ( in
the mean time we sent them several crash dumps showing the reservation
conflict and applied every single patch going per Sun's request ).

As a result, looking to either use VxVM or possibly Sun Cluster (
which appears to use different types of SCSI reservations to SVM on
it's own - haven't tried it out though and again the QLogic cards
would probably make it an unsupportable configuration for Sun Cluster
)

Thanks to kendall lloyd for pointing out that once you put a LUN into
a diskset, mpxio will always treat it as active/passive regardless of
whether the array serving the LUN is active/active capable.

On 27/02/06, Tim Wright <tim.dormouse@googlemail.com> wrote:
> Hi
>
> Have a V240 running S8 with SVM 4.2.1 ( +latest patch ) and SFK 4.4.8
> ( again with latest patches ).
>
> Have 2 QLogic QLA2340 HBAs connected via a SAN fabric ( actually 2
> independent fabrics, one per HBA ) to 2 controller channels on a
> transtec 6600 fibre array.
>
> Using SUNWqlc ( leadville ) drivers to manage the HBAs ( also tried
> QLogic's own drivers and fail over but they give the same problem -
> QLogic support then recommended that we don't use their drivers apart
> from on Linux as they are not developing for other OS's ).
>
> Using mpxio for the luns on the array which load balances across both HBAs.
>
> Normal course of events, everything's fine, if disable one of the HBAs
> ( by disabling port on switch ) remaining path is ok and re-enabling
> the HBA path works as well.
>
> However, if out any of the array LUNS into a metaset, then if the path
> through which SVM issues the SCSI reservation on the LUN is disabled,
> it tries to reissue the SCSI reservation through the other path finds
> a conflict and consequently the host panics ( as would be expected )
>
> I have searched all the Sun documentation I can find as well as
> googled extensively but I cannot find anything which mirrors my
> situation exactly - has anyone had similar issues and if so did they
> ever resolve them.
>
>
> tim
>
> PS. using metasets because eventually plan to share these LUNs between
> two or more hosts but currently just testing with one host ( one of
> the docs I was reading from Sun even suggested use metasets even if
> not shared LUNs )
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Sun May 14 11:41:08 2006

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:58 EST